The concept of E-learning has gained popularity among universities and students in recent years as E-learning education platforms can record student learning behaviours in its many forms to recognise and analyse student learning styles. However, it is known that the challenges this brings in monitoring student engagement with course material can be considerable and variable between STEM (Science, Technology, Engineering, Math) and Social Sciences courses. This paper applies a graph-based community detection method that integrates the cumulative actions of a student with the Virtual Learning Environment (VLE), through a preprocessing technique, facilitating deeper analysis of student performance using the OULAD dataset1. Our findings reveal that this method is trustworthy, and we show that it outperforms traditional classification and clustering methods and achieves superior accuracy in evaluating and predicting academic outcomes—encompassing both formative assignments and terminal assessments. Moreover, this method uncovers variations in learning styles among students in STEM and Social Sciences, indicating commonalities and diversity of the learning approach for the different types of classes, in the same E-learning system.